Probabilistic classification of forest degradation by selective logging and fires in the Amazon based on Sentinel-2 data

Ekena Rangel Pinage,  Oregon State University,  ekenapinage@hotmail.com (Presenter)
Michael Keller,  USDA Forest Service,  mkeller.co2@gmail.com
Marcos Longo,  Lawrence Berkeley National Laboratory,  mdplongo@gmail.com
Paul Duffy,  Neptune, Inc.,  paul.duffy@neptuneinc.org
Ovidiu Csillik,  Jet Propsulsion Laboratory,  ovidiu.csillik@gmail.com
Doug Anderson,  Neptune, Inc.,  danderson@neptuneinc.org

Tropical forests worldwide have been transformed by deforestation, logging, fragmentation, and fires. While estimates of deforested areas are relatively reliable in the Amazon, there is great divergence in forest degradation estimates due to the intrinsic characteristics of degradation events, their distribution in space and time, and the challenges to detecting them with satellite observations. Existing forest degradation classification efforts have considerable detection uncertainty, which generally has not been considered when accounting for carbon emissions from degraded areas. Using textural metrics computed from Sentinel-2 images at 10-m resolution and then aggregated to a 500-m grid size, we implemented a probabilistic classification of forests degraded by selective logging and fires in six sites across the Brazilian Amazon. Our overall accuracy was 0.75 when including degradation events up to 5 years prior to the reference year. Fires were well detected by our classification approach (sensitivity = 0.83), whereas there was greater confusion between intact and logged forests (sensitivity: 0.73 and 0.72, respectively). We are refining the model to include shorter intervals of forest degradation, from which we expect increased accuracy. Subsequently, we will extend this approach across the Amazon Arc of Deforestation, a region that concentrates most forest disturbances in the Brazilian Amazon. Reliable classification of forest degradation area is critical for the quantification of carbon losses and gains in the Amazon region.

Poster: Poster_Rangel_Pinage_2-53_60_35.pdf 

Poster Location ID: 2-53

Presentation Type: Poster

Session: Poster Session 2

Session Date: Wed (May 10) 5:15-7:15 PM

CCE Program: LCLUC